Watson promises 3,100-site people search, but its public search code returns a demo

A free app promising to investigate somebody else deserves a small investigation of its own.

Watson’s README advertises a Windows desktop app that takes a nickname, email address, phone number or website and builds a report from more than 3,100 public pages. It promises a readable answer in about 30 seconds, no account and no paid API keys. That would be a handy alternative to a weekend spent playing browser-tab detective.

The public source, however, describes a much earlier stage. Its developer table points directly to src/data/mock.ts as the place containing example data and the function where real lookups should be connected. It is quite an important footnote for software whose main attraction is doing lookups.

The results screen imports searchProfiles from that mock module and calls it with the search query. The function returns a predefined profile array if the trimmed, lowercased input matches the built-in example nickname, johndoe_92. Other inputs get an empty array.

There is no network lookup in that function. The code’s introductory comment calls the module “Built-in example data,” and its explanation says the examples remain until real lookups are connected. The included examples list accounts, case relationships and even tool-run counts. Those are interface fixtures. A screen full of numbers can look wonderfully industrious while doing no investigating at all (excellent posture, though).

That finding applies to the inspected public source path. It does not establish what a separately distributed Windows binary contains, whether another branch has a working backend or why the README’s marketing and implementation differ. The binary was not installed or tested here. Accusing somebody of a motive would require evidence that this repository does not supply.

It does change how a reader should interpret the demo. A match for the sample nickname demonstrates that the screen can render its example data. A blank response for your own name demonstrates the function’s other branch. Neither result establishes what the internet knows about you.

The README also describes green same-person matches and gray nickname-only matches, plus PDF export and local history. Those are claims about the advertised product. Before relying on that product to assess a seller, a recruiter or your own public footprint, an actual lookup needs to be demonstrated with independently checkable output. Pretty colors do not perform identity verification.

There is a useful project idea underneath the mismatch: one interface that distinguishes a reused nickname from corroborated identity could save people from very bad assumptions. The next meaningful milestone in this public implementation is connecting real queries to real evidence. Until that path exists, the detective has a desk and a very convincing sample case.

The case file is prefilled.